我目前正在使用tensorflow和cifar10开发一个模型。
输入尺寸是从cifar10加载的。
(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.cifar10.load_data()
train_images, test_images = train_images / 255.0, test_images / 255.0
nb_classes = len(numpy.unique(train_labels))
train_labels = tf.one_hot(train_labels, nb_classes)
test_labels = tf.one_hot(test_labels, nb_classes)
输入形状为(10000,32,32,3)，测试形状为(10000,1,10)。
我收到一个错误与我的模型的层的最后一行。
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 32, 32, 3)] 0
__________________________________________________________________________________________________
conv2d (Conv2D) (None, 32, 32, 32) 2432 input_1[0][0]
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 32, 32, 32) 25632 conv2d[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 32, 32, 32) 25632 conv2d_1[0][0]
__________________________________________________________________________________________________
add (Add) (None, 32, 32, 32) 0 conv2d_2[0][0]
conv2d[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 32, 32, 64) 51264 add[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 32, 32, 64) 102464 conv2d_4[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 32, 32, 64) 51264 conv2d[0][0]
__________________________________________________________________________________________________
add_1 (Add) (None, 32, 32, 64) 0 conv2d_5[0][0]
conv2d_3[0][0]
__________________________________________________________________________________________________
flatten (Flatten) (None, 65536) 0 add_1[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, 512) 33554944 flatten[0][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 10) 5130 dense[0][0]
==================================================================================================
我得到错误信息"ValueError:形状(None, 1,10)和(None, 10)是不兼容的。
我不知道如何解决这个错误。
这是因为CIFAR-10标签嵌套在形状(None, 1)而不是(None)中。
>>> a = [[0], [1], [2], [3]]
>>> tf.one_hot(a, 4)
<tf.Tensor: shape=(4, 1, 4), dtype=float32, numpy=
array([[[1., 0., 0., 0.]],
[[0., 1., 0., 0.]],
[[0., 0., 1., 0.]],
[[0., 0., 0., 1.]]], dtype=float32)>
>>> b = [0, 1, 2, 3]
>>> tf.one_hot(a, 4)
<tf.Tensor: shape=(4, 4), dtype=float32, numpy=
array([[1., 0., 0., 0.],
[0., 1., 0., 0.],
[0., 0., 1., 0.],
[0., 0., 0., 1.]], dtype=float32)>
您可以看到嵌套标签生成嵌套的单热编码标签。在您的情况下，您可以只在单热编码前flatten或reshape标记:
...
nb_classes = len(numpy.unique(train_labels))
train_labels = train_labels.flatten()
test_labels = test_labels.flatten()
train_labels = tf.one_hot(train_labels, nb_classes)
test_labels = tf.one_hot(test_labels, nb_classes)
现在标签的形状应该对应于模型的输出形状。
你的train_labels的原始形状是(50000,1)，你需要删除1使用numpy.squeeze)，这将使形状(50000)相同的处理test_labels
(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.cifar10.load_data()
train_images, test_images = train_images / 255.0, test_images / 255.0
nb_classes = len(numpy.unique(train_labels))
test_labels=numpy.squeeze(test_labels)
train_labels=numpy.squeeze(train_labels)
train_labels = tf.one_hot(train_labels, nb_classes)
test_labels = tf.one_hot(test_labels, nb_classes)